-
Notifications
You must be signed in to change notification settings - Fork 5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Support optionally using karras sigmas in the k_diffusion wrapper pipeline #1633
Comments
I will try to allocate some time in the next weeks to look into it (can't promise to get it done soon though). Just wondering: do we have some results/benchmarks to know whether karras sigmas work better ? |
I don't think they necessarily work better. It just works differently and the community enjoys having the possibility of having this variations for artistic purposes when using tools that support it such as AUTO1111 and any other based on k_diffusion so I think it would be great for diffusers to also enable such variations. But I can't say it works better (some people believe so subjectively) |
In case one is interested in taking a look? @williamberman @yiyixuxu @pcuenca @patil-suraj |
Here a solution that should work well: #1887 (comment) . Hope to soon find time to implement it :-) |
Did you add the karras sigmas to diffusers's scheduler? I want to use DPM++ 2M karras but I only find DPMSolver multi-step. |
Not yet @ym-ctw - we'd be more than happy to review a PR though :-) |
Thanks, I really hope you could add karras especially in DPM++ series soon. Because now much more new model are using 'DPM++ 2M karras' to get much better results than 'eular a' and others schedulers, most of these models are from 'civitai.com.' |
Is your feature request related to a problem? Please describe.
K-Diffusion enables users to generate sigmas with
k_diffusion.sampling.get_sigmas_karras
.diffusers
currently do not support using those to generate the sigmas with its k_diffusion wrapperDescribe the solution you'd like
Enable users to optionally generate karras sigmas similarly to how AUTO1111 does it https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/44c46f0ed395967cd3830dd481a2db759fda5b3b/modules/sd_samplers.py#L453
The text was updated successfully, but these errors were encountered: